Sarcopenia measured by artificial intelligence as a predictor of overall survival in localized bladder cancer, a multicenter study.

BACKGROUND

Sarcopenia, characterized by a loss of skeletal muscle mass and function, is a poor prognosis risk factor in various cancers. Diagnosis typically involves quantifying skeletal muscle area at the L3 vertebral level (SMA) using CT imaging, however a universally accepted definition of sarcopenia does not exist.

METHODS

In a retrospective, multicenter study, we analyzed data from 87 muscle-invasive bladder cancer patients who received neoadjuvant chemotherapy followed by radical cystectomy. Artificial intelligence (AI) was used to evaluate CT scans obtained before chemotherapy (BC) and before surgery (BS), focusing on the L3 vertebral level. Sarcopenia was defined using four distinct criteria from the existing literature. The primary objective of this study was to assess the reliability of AI-based versus manual measurements of sarcopenia while secondary endpoints included, overall survival (OS), progression-free survival (PFS), and the impact of undernutrition and neutrophil-to-lymphocyte ratio (NLR) on OS.

RESULTS

87 patients were included in the final analysis. AI-based SMA measurements were highly correlated with manual measurements (r = 0.95; p < 0.001). Sarcopenia BC and BS, as defined by the Pardo criteria, was significantly associated with poorer OS (Prado BC: HR 2.26, 95% CI 1.05-4.89, p = 0.04 and Prado BS: HR 2.10, 95% CI 1.01-4.37, p = 0.048). Sarcopenia BS, defined by Caan was also associated with worse OS (HR 2.15, HR 1.02-4.56, p = 0.04). Both NLR ≤ 4 BS and undernutrition were strongly associated with reduced OS (HR 3.33; 1.49-7.46, p = 0.003 BC and HR 2.85; 95% CI 1.3-6.3, p = 0.003, respectively).

CONCLUSION

AI-based assessment of sarcopenia is feasible, reliable, and reproducible. Sarcopenia according to Prado's criteria is associated with mortality in bladder cancer. Further multivariable validation in larger patient populations is needed to determine the prognostic value of AI-based body composition analysis.